LAMPSS: Discovery of Metal-Poor Stars in the Galactic Halo with the Cahk ﬁlter on CFHT Megacam ?

NLUAstro 1, 75-90 (2019) Notices of Lancaster University Astrophysics PHYS369: Astrophysics Group project

LAMPSS: Discovery of Metal-Poor Stars in the Galactic Halo with the CaHK ﬁlter on CFHT MegaCam ?

Eleanor Worrell1, Niam Patel1, Karolina Wresilo1, Sam Dicker1, Olivia Bray1 Jack Bagness1 & David Sobral1† 1 Department of Physics, Lancaster University, Lancaster, LA1 4YB, UK

Accepted 31 May 2019. Received 31 May 2019; in original form 25 March 2019

ABSTRACT We present the Lancaster Astrophysics Metal Poor Star Search (LAMPSS) in the Milky Way halo, conducted using the metallicity-sensitive Ca H & K lines. We use the CaHK filter of the MegaPrime/MegaCam on the Canada-France-Hawaii Telescope (CFHT) to survey an area of 1.010deg2 over the COSMOS field. We combine our new CaHK data with broadband optical and near-infrared photometry of COSMOS to recover stars down to mHK ∼ 26. After removing galaxies, we find 70% completion and 30% contamination for our remaining 1772 stars. By exploring a range of spectral types and metallicities, we derive metallicity-sensitive colour-colour diagrams which we use to isolate different spectral types and metallicities (−5 ≤ [Fe/H ≤ 0). From these, we identify 16 potentially extremely metal-poor stars. One of which, LAMPS 1229, we predict to have [Fe/H] ∼ −5.0 at a distance of (78.21 ± 9.37) kpc. We construct a Metallicity Distribution Function for our sample of metal-poor stars, finding an expected sharp decrease in count as [Fe/H] decreases. Key words: Galaxy: Halo - Stars: Population III - stars: abundances - Galaxy: evolution - Galaxy: Archaeology

1 INTRODUCTION main sequence today. The Caterpillar simulation suite pre- dicts that a Milky Way-like host galaxy would have many The ΛCDM model suggests the Universe can be traced back Population III formation sites at z ∼ 25 (e.g. Griffen et al. to a singularity 13.8 Gyrs ago (Lemaˆıtre 1927; Macaulay 2018), meaning there is the exciting possibility that Pop- et al. 2018). Several hundred million years after, the first ulation III stars may have formed in, and still be present stars formed. These Population III stars would have been in, our galactic halo. Direct observation of a Population III composed entirely of primordial gas (hydrogen, helium and star would be of great significance for various areas of astro- trace amounts of beryllium and lithium). The existence of physics and cosmology. Firstly, it allows unparalleled access Population III stars is expected, as heavy elements present in to nucleosynthesis within the very first generation of stars the Universe can only be synthesised in the interior of stars. which can help validate current knowledge of stellar nucle- It has long been thought that these first generation stellar osynthesis. It has been predicted, via simulations, that a Populations would be exclusively high mass stars (∼100M ) significant portion of galaxies are “Population III - bright” (e.g. Bromm et al. 1999; Schaerer 2002). However, more re- immediately prior to the epoch of reionisation (Sarmento cently it has been simulated that the metal-free clouds of et al. 2018). Therefore, understanding Population III stars the early Universe would allow protostars of 0.1M (e.g. Be- is a crucial step in fully understanding the epoch of reion- cerra et al. 2015; Stacy et al. 2016) and that the primordial isation in the Universe (Koh & Wise 2018). Furthermore, IMF may extend below 0.8M (e.g. Hartwig et al. 2015). observing a Population III star would constrain the IMF of At those masses, stellar lifetimes are greater than cosmic Population III stars which is currently uncertain. It would time so Population III stars could still be observable on the also provide insight into the formation of the proto-Milky Way as well as any accretion events in our galaxy’s history ? Based on Cosmic Evolution Survey (COSMOS) data, CaHK (e.g. Sestito et al. 2019). CFHT data (programs 15BC10, 16A038 & 17BO54 PI: Sobral), theoretical spectral models provided by POLLUX, and images Metallicity is defined as the mass fraction, Z, of ele- from the HST. ments heavier than helium in the star. Therefore, Population † PHYS369 supervisor. III stars should theoretically have zero metallicity. Further-

c 2019 The Authors 76 Worrell et al. (LAMPSS) more, massive Population III stars can go supernovae within The methods used to reduce the initial LAMPSS catalogue, just a few million years, releasing metals into the Universe as well as the construction of metallicity-sensitive colour- for the first time. These metals are then available to go on to colour diagrams, are both presented in §4. In §5 we present form early Population II stars which should also be very low the 16 EMP candidates identified and give stellar number in metals. Metallicity is commonly quantified by the ratio of densities per spectral type per metallicity for the region sur- iron to hydrogen within a star, [Fe/H]. [Fe/H] represents a veyed. Throughout this paper we use AB magnitudes (Oke logarithmic comparison of this ratio with respect to the solar & Gunn 1983). ratio. Metal-poor stars are classified as follows; extremely, ultra, and hyper metal-poor (EMP, UMP and HMP hereafter) stars with [Fe/H]< -3.0, [Fe/H] < -4.0, and [Fe/H]< 2 THEORETICAL BACKGROUND -5.0, respectively (Beers & Christlieb 2005). Several metal-poor star studies have been conducted. 2.1 Composition of Population III Stars One of the most recent is the Pristine survey (Starken- Once the temperature of the Universe dropped to 109 K, fu- burg et al. 2017), a narrow-band survey which used the sion of hydrogen, deuterium, tritium and helium could occur metallicity-sensitive Ca H & K lines (CaHK hereafter), via the processes shown in Equation1 1. Heavier elemental which lie at 3968.47A˚ and 3933.66A˚ respectively. The isotopes were unable to form in the recombination epoch2. success and potential of the survey at identifying EMPs Helium has too great a Coulomb barrier compared to the is unprecedented, with the survey having already con- kinetic energy available (which drops with temperature as firmed one of the most metal-poor stars found, Pristine the Universe expands), for helium fusion to occur. 221.8781+9.7844 with [Fe/H] = -4.66 ± 0.13 (Starkenburg + 2 et al. 2018). Broad-band photometric surveys can also iden- p + n H + γ tify metal-poor stars, such as SDSS J102915+172927, the 2H +2 H 3H + p (1) Caffau dwarf star, with [Fe/H] = -4.5 (Caffau et al. 2011). 3 2 4 The most iron deficient star is the Skymapper Southern Sky H + H He + n Survey star SMSS J031300.36 - 670839.3 which has [Fe/H] Another factor that inhibits production of elements < -6.53 (Keller et al. 2014; Nordlander et al. 2017). Re- heavier than helium before first generation stars, is the sup- cently, restrictions in metal-poor star studies due to small pression of neutron capture due to neutron decay and the sample size have been lifted as many more metal-poor stars expansion of the Universe. Neutron capture, which is the are found, for example, the hundreds of new metal-poor first process shown in Equation1, is a necessary step in stars identified by the R-process Alliance from RAVE data the fusion of helium-4, as shown above in Equation1. The (Sakari et al. 2018; Hansen et al. 2018). Furthermore, fu- half-life of a free neutron is approximately 10 minutes, there- ture Population III searches have been proposed, where pair- fore any free neutron would quickly decay into a proton and instability supernovae could be used as a high redshift probe electron, making the chance of a stable helium nuclei signif- (e.g. Hartwig et al. 2018). icantly less likely3. After inflation, the rate of the expansion Another major problem in the search for primordial of the Universe has been accelerating (e.g. Scharping 2017), stars thus far, is that the majority of metal-poor stars are decreasing the density of matter with time from the recom- carbon enhanced (CEMP), such as the stars G64-12 and bination epoch. This meant neutron capture, and thereby G64-37 (Placco et al. 2016a). However, UMP stars without production of heavier elements, would get exponentially less enhanced carbon abundance have been found, such as Pris- likely as time advances until other methods of nucleosynthe- tine 221.8781+9.7844 and the Caffau star. Additionally, it sis are viable. has been found that in the Sagittarius dwarf galaxy there Elements heavier than helium can only be generated exists a very metal-poor stellar population and that out of through stellar nucleosynthesis and supernovae (Ryan & four newly identified stars, none are CEMP stars (Chiti & Norton 2010), where the temperature is such that particles Frebel 2019). Further work has suggested that multiple stel- have sufficient energy to overcome the Coulomb barrier of lar progenitors are required at [Fe/H] -4.0 (e.g. Placco et al. the helium nuclei. As a result, the primordial Population III 2016b), meaning multiple formation routes for UMP stars. stars are composed of only hydrogen and helium with trace Numerical simulations suggest that the halo is the most amounts of lithium. likely place to find Population III survivors (e.g. Ishiyama et al. 2016; Magg et al. 2018). It follows that the Pop- ulation III stars most likely to have escaped enrichment 2.2 Why are Population III stars elusive? from supernovae winds will be towards the outskirts of the halo. In this report we present one of the first attempts to There are several complications when searching for Popula- probe the extra-galactic field for metal-poor stars, using the tion III stars. The age of any surviving Population III stars narrow-band CaHK filter of the MegaPrime/MegaCam on is of the order of cosmic time, so by looking at Table1, it is the Canada-France-Hawaii Telescope (CFHT) in combina- almost certain that the majority of Population III stars will tion with the broad-band optical and near-infrared photom- have long since left the main sequence and completed their etry available in the COSMOS field. We survey 1.010 deg2 of sky in the northern hemisphere to ∼26 mag, 5 magnitudes 1 http://burro.case.edu/Academics/Astr222/Cosmo/Early/ deeper than the aforementioned Pristine survey. bbn.html This report is organised as follows. §2 gives background 2 http://abyss.uoregon.edu/~js/ast123/lectures/lec21. on Population III stars. In §3 we present the sample, drawn html from the COSMOS catalogue with CaHK data from CFHT. 3 https://wmap.gsfc.nasa.gov/Universe/bb_tests_ele.html

NLUAstro 1, 75-90 (2019) LAMPSS: Lancaster Astrophysics Metal Poor Star Search 77

Spectral Type Colour Temperature (K) Main Sequence Lifetime lines rather than the absorption lines associated with iron. O Blue >30,000 <5 Myrs B Blue 11,000 - 30,000 5 - 800 Myrs Fe absorption lines are usually small and are found across A Light Blue 7,500 - 11,000 800 Myrs - 3 Gyrs a large range of wavelengths, making them extremely hard F White 6,000 - 7,500 3 - 8 Gyrs to measure without high resolution spectroscopy. This issue G Yellow 5,000 - 6,000 8 - 18 Gyrs is amplified for metal-poor stars that are in the stellar halo K Orange 3,500 - 5,000 18 - 50 Gyrs M Red <3,500 >50 Gyrs (∼ 10 kpc away). CaHK lines can be used to infer stellar metallicity in Table 1. Main sequence lifetimes of spectral types, calculated in place of iron, as the abundance of calcium typically traces Appendix ??. that of iron in a star. The abundances are similar as they are both synthesised by the r-process during supernovae of massive stars or white dwarfs. The r-process, or rapid pro- evolution. Especially since it is predicted that the majority cess, invovles the capturing of neutrons by seed nuclei over of Population III stars were much larger than later stellar a timescale such that the nuclei cannot undergo nuclear de- generations (M 100 M )(Stacy et al. 2016), which have & cay before being bombarded by another neutron (Arnould lifetimes of ∼10 Myrs (e.g. Weidner & Vink 2010). Hence, et al. 2007). Therefore, the ratio [Ca/H] will be reflective of they now would be observed as remnants such as neutron [Fe/H]. Figure1 shows the variation of the CaHK absorp- stars or black holes. The original composition of these rem- tion profile with [Fe/H] for a Sun-like star. While the depth nants is increasingly difficult to determine due to the loss of the lines does not noticeably change, the breadth of the or consumption of internal elements during the process of absorption lines increases with increasing [Fe/H], thereby stellar evolution (e.g. Laughlin et al. 1997). making stars look fainter at these wavelengths. As well as There is also the potential that low mass Population singly-ionised calcium, there are numerous other elements III stars have become contaminated with heavier elements present in an atmosphere of a star such as the Sun. This when moving through the ISM, altering the composition of results in many absorption lines spread over a wide range the outer layers. Population III stars may undergo a pro- of wavelengths, which can be best observed in the [Fe/H]=0 cess known as self-contamination through the dredge up of line of Figure1. products of stellar nucleosynthesis in the core, causing the As a consequence, metallicity directly affects the appar- appearance of a higher metallicity, typically associated with ent colour of a star. Since metals preferentially absorb light Population II stars4. of shorter wavelengths, metal-rich stars appear redder than their metal-poor partners at a fixed surface temperature. 5 2.3 Metallicity Determination and Fraunhofer This effect is called line blanketing . Lines If metals are present in the stellar atmosphere, it is possible 2.4 What spectral Type would Population III to determine the stars’ metallicity through analysis of their Survivors Be? spectra. Elements display their own characteristic absorption lines and hence they can be easily identified. As photons A low mass Population III survivor would have an age of travel through the stellar envelope they can be captured by the order of the age of the universe. From the main se- atoms if the photon has energy equal to the difference in en- quence lifetimes shown in Table1, it is clear that popu- ergy between 2 energy states of that atom. When this light lation III survivors must be G or K type stars. Based on is re-emitted in all directions, the overall light intensity at simulations of the primordial halo evolution, it was thought that particular wavelength decreases causing a distinctive stars with mass greater than 100 M were more favourable absorption lines. Stellar metallicity is typically defined by (e.g. Bromm et al. 2002). It was argued that primordial gas [Fe/H]; a logarithmic ratio of the abundance of iron relative can only perform continued gravitational collapse and frag- to hydrogen. mentation, and therefore star formation, if the gas can effectively radiate away the gravitational binding energy that is released in this process. This process is only possible at NFe NFe [Fe/H] = log − log (2) 6 10 10 gas cloud masses above 10 M , which under collapse only NH ∗ NH produces large stars. However, in recent years, as computa- Here, NFe and NH represent the number densities of iron tional capabilities and understanding of fragmentation have and hydrogen atoms, respectively. The logarithmic metallic- improved, Stacy et al.(2016) have provided greater simula- ity of a star is compared to that of the Sun, such that the tion resolution of mini-halos showing protostars form with more negative the ratio [Fe/H], the lower the abundance of masses between 0.1 to 1000M , mainly due to fragmenta- metals within in that star. [Fe/H]= 0 indicates that a star tion of protostellar discs, therefore suggesting that G and has solar metallicity. K type stars were feasible in the early Universe. M spectral Usually this metallicities are measured by comparison type stars would form in the early Universe, but would be of photometric and spectroscopic measurements from a star. heavily suppressed. The problem with this simulation is that Fraunhofer CaHK lines are a signature of singly ionised cal- it only produces protostars, therefore it is assumed that all cium being present in the atmosphere of a star. The absorp- these protostars would go on to perform nucleosynthesis and tion occurs at 3968.47 for the H line and at 3933.66 for the the final stellar mass will be much larger. K line. Often, it is more feasible to look for CaHK absorption

4 http://astronomy.swin.edu.au/cosmos/P/Population+III 5 https://iopscience.iop.org/article/10.1086/128651/pdf

NLUAstro 1, 75-90 (2019) 78 Worrell et al. (LAMPSS)

Figure 1. Plot to show the effect of varying [Fe/H] on the CaHK lines (at 3968.47 and 3933.66 respectively) on a G type star with temperature 5750K. Note the significant effect of metallicity on the CaHK absorption features, making it a very sensitive [Fe/H] indicator.

2.5 Why Search for Metal-poor Stars in the Halo? the merger rate was incredibly high in the early Universe and lastly the high temperature of the CMB (T > 30K) limits The halo of a galaxy is a broad, roughly spherical structure the cooling of the gas making it difficult to form stars. Once which extends beyond the visible edge of the galaxy. Includ- the Universe cools, and finally allows the formation of larger ing dark matter, the halo contains the vast majority of a stellar structures, two types of GCs arise, metal poor and galaxies mass and, in the case of the Milky Way, is approx- metal rich GCs based on their location of formation. GCs imately 100 kpc from the Earth6. The location of Popula- formed inside the radius of the host primary galaxy are gen- tion III stars would likely be in globular clusters close to the erally more metal rich whereas GCs captured from outside galactic nucleus, as this appears to be the edge at which Pop- the primary galaxy are more metal poor (Forbes et al. 2018). ulation II stars are observed. However looking towards the The position of the ‘beginning’ of the halo can be estimated galactic nucleus is like looking through fog. In comparison, by finding the distance at which the number of stars in a looking towards the galactic halo bears less contamination data set abruptly drops off. Current estimates for the dis- of the data set with Population I stars. In addition, stars tance of the Milky Way’s stellar halo is at approximately found in the nucleus have a high chance of being contami- 30 kpc,7 to around 200 kpc8. nated by metals from stellar winds, and therefore lose their metal-poor status. Some hypotheses suggest that the formation processes of Globular Clusters (GC) mean Population III are present 2.6 Background Galaxies in GCs, although there have been many theorised postu- While surveying the sky, and particularly at very low fluxes, lates based upon GC formation, and not all of them agree number counts become dominated by distant galaxies that on one hypothesis. One such postulate, proposed by Trenti may resemble stars. This leads to the need to carefully dis- et al.(2015), suggests that mergers of gas-rich, but star free, tinguish between stars and galaxies. Edwin Hubble, in 1929, mini-halos can cause instantaneous compression and form while measuring distances to galaxies, concluded that the central stellar clusters. These would eventually be stripped Universe, and any space where objects are not bounded by of their dark matter envelope and become the GCs with an- electromagnetic or gravitational forces, was expanding at a cient stellar populations that we see today. This postulate velocity directly proportional to the distance the object is seems likely due to fundamental physics it is based upon in this high redshift era. Firstly the actual formation of large spiral galaxies in this period was significantly low, secondly 7 https://imagine.gsfc.nasa.gov/features/cosmic/ milkyway_info.html 6 https://www.cfa.harvard.edu/research/oir/ 8 https://ned.ipac.caltech.edu/level5/Ashman2/Ashman3. milky-way-halo html

NLUAstro 1, 75-90 (2019) LAMPSS: Lancaster Astrophysics Metal Poor Star Search 79 from us (e.g. Hubble 1929). This is shown by equation3,

~v ∝ d~ ~v = Hd~ (3) where v is recessional velocity, d is distance from Earth and H is the Hubble Constant. At large recessional velocities, the redshift, z, must take into account relativistic corrections. When the object travels directly away from the observer, the redshift is given by equation4( c is the speed of light in a vacuum).

r c + v r c + Hd z = − 1 ⇒ z = − 1 (4) c − v c − Hd A star within our halo will have z ≈ 0, although there may be some redshift due to a stars rotation about the galactic centre. Galaxies outside the Local Group will have con- siderably larger redshifts.

2.7 AB Magnitude System and Colour-Colour Diagrams Figure 2. Normalised transmission profiles for various filters used AB magnitudes (Oke 1974) are defined as follows, in this study.

mAB = −2.5 log10 fν − 48.60 (5) Filter ∆λ (A˚) λ (A˚) where m is the AB magnitude and f is the flux density in eff AB ν i 1168.1 7641.3 units of erg s−1 cm−2 Hz−1. Flux density (defined per unit g 1304.2 4912.8 wavelength, f ) can be converted into flux density (flux per λ CaHK 94.2 3953.4 unit frequency) using equation6, v 972/4 5505.9 λ2 r 1092.9 6251.1 f = f (6) ν c λ u 528.7 3835.2 where c is the speed of light and λ is the wavelength. Table 2. Summary of the bandwidths (∆λ) and effective trans- Colour-colour diagrams are created using different com- mission wavelengths (λeff ) of each filter obtained from the filter binations of magnitudes in different bands as the axis. An profiles. example of a colour index, (g − i), where g and i are the magnitudes in the green and the infrared filter respectively. Stars are effectively black bodies, so a colour index is directly which added optical and near-infrared magnitudes, such as related to properties of astronomical objects, such as their u, g, r, i, v, j, z and k and their associated errors. The fil- intrinsic colour or temperature. Therefore the positioning of ter profiles of u, g, r, i, v and CaHK are plotted in Figure data-points on colour-colour diagrams can, for example, in- 2 The bandwidths and effective wavelengths of u, g, r, i, v dicate the relative metallicities of corresponding stars. Since and CaHK are displayed in Table2. metal-rich stars appear redder, we can expect their (g − i) A record of the subsequent cuts made to the initial indices to be more positive than those of their metal-poor LAMPSS catalogue are shown in Table3, and a descrip- counterparts. tion of how these cuts are made is given in the relevant subsections of §3.

3 DATA AND SAMPLE 3.2 Magnitude Cuts 3.1 Initial Catalogue In order to mitigate large uncertainties in the metallicity and Our dataset used CaHK observations conducted with the (g − i) conditions, we excluded sources that are too faint by MegaPrime/MegaCam on the CFHT (programs 15BC10, placing a magnitude limit on several bands. This also re- 16A038 & 17BO54 PI: David Sobral)9, together with masks moved sources which have undeﬁned magnitudes in certain previously designed by D.Sobral, to produce a CaHK cata- bands (which appear blank or as 99.9 in the catalogue). The logue of sources using . The 200 diameter aper- SExtractor limits used are shown in Table4. For bands imported from ture used for CaHK magnitudes found 62,652 sources, ex- the COSMOS catalogue we used cuts given in Capak et al. tending to a depth of m = 26. HK (2007). For CaHK magnitudes we simply devised our own This CaHK catalogue was then combined with broad- magnitude limit, by observing at what magnitude the counts band COSMOS photometry data (Scoville et al. 2007), in Figure3 begin to fall oﬀ ( mHK = 26). By a miscommuni- cation, mHK = 25.3 was used for volume calculations (§4.7). 9 Information on these programs can be found at: However, since limiting at mHK = 26 only removes 0.973% http://www.cadc-ccda.hia-iha.nrc-cnrc.gc.ca/en/search/ of the catalogue and due to time constraints, we decided the ?collection=CFHT&noexec=true incompleteness this created was negligible.

NLUAstro 1, 75-90 (2019) 80 Worrell et al. (LAMPSS)

Catalogue Described Sources Notes Version in Section n/a 2,017,800 Full COSMOS photometry n/a 0.0 62,652 Merge COSMOS and CaHK data 3.1 1.0 1,930 Magnitude and EGO cuts 3.2/3.3 1.1 1,930 Added internal ID to each source n/a 1.2 1,930 Added distances to each source 4.6 1.3 1,930 Applied metallicity conditions 4.4

Table 3. Walkthrough of all major catalogue versions, number of sources within each version and notes on cuts made.

+ + ∗ + found by plotting (J − K) against (B − u) (see Figure4), i r u VJ g BJ mHK and gave the following criteria for stars: 26.2 26.8 26.4 26.6 27.0 27.3 26.0

Table 4. Magnitude cuts performed on our catalogue, all ex- (J − K) < −0.6143(B − u) − 1.6121 (7) cept mHK are given magnitude limits for COSMOS Capak et al. (2007). (B − u) < −0.674 (8) There were two conditions, as the contamination seemed to increase as (B-u) increased as shown in ﬁgure4. Completion and contamination are deﬁned below by equations9 and 10. The (J − K) against (B − u) criteria returned a contamina- 12000 tion of 30% and a completion of 70%.

10000 N Completion = stars kept (9) Nstars total 8000 N Contamination = galaxies kept (10) Ngalaxies kept + Nstars kept 6000 Count

4000 4 METHODOLOGY

2000 Follwing our sample reduction, we were left with 1930 sources which we classified as stars. However, since our star- galaxy separation has 30% contamination, it is likely that 0 16 18 20 22 24 26 28 many of these 1930 sources will be galaxies. Due to time con- CaHK Apparent Magnitude straints, we could only manually check the nature of sources with low metallicty using HST imagery (§4.5). In this section we describe our methods to classify these stars by spectral Figure 3. Histogram of CaHK magnitudes of sources through a type and metallicity, as well as calculate distances to the 00 3 diameter aperture. The vertical line shows the magnitude cut stars. applied to the catalogue at mHK = 26, removing 0.973% of the catalogue. 4.1 Magnitude Calculations 3.3 Star-Galaxy Separation In order to interpret observed magnitudes, we analysed a range of spectra of stars of known temperatures, spectral The primary purpose of the COSMOS survey was to observe types and metallicities. This was necessary to compute mag- galaxies and their evolution over time, as a consequence, the nitudes in a range of filters and construct a colour diagram. majority of sources in the LAMPSS catalogue are Extra- The effect of temperature and metallicity could then be seen Galactic Objects (EGOs). These EGOs had to be removed through carefully choosing appropriate colour combinations before searching for metal-poor stars within our halo. to be plotted on the y- and x- axes. Stellar spectra, obtained J and K are two infrared bands between 1.1 to 1.4 µm from the POLLUX database10, and filter profiles of differ- and 2.0 to 2.4 µm respectively. Sources at a higher redshift ent filters of the MegaCam/MegaPrime on the CFHT were have a larger (J−K), than those at smaller redshifts, as such, interpolated and re-expressed as a continuous function of (J − K)-colour diagrams are an effective way to seperate wavelength in a Python script. An example of an interpo- stars and galaxies. lated spectrum can be found in Figure5. Using TOPCAT subsets, we defined sources within our The filter profiles, shown in Figure2, were normalised, catalogue with zero spectroscopic redshift (z spec = 0) as such that they consisted of relative light transmission, rang- stars and sources with 0.1 ≤ z ≤ 8 as galaxies. We inves- ing between 0 and 1, as a function of wavelength. The band tigated plots of (J − K) against other colour indices were made in order to determine the cleanest cut, all plots tri- alled are in Table3 in Appendix ??. The cleanest cut was 10 http://npollux.lupm.univ-montp2.fr/

NLUAstro 1, 75-90 (2019) LAMPSS: Lancaster Astrophysics Metal Poor Star Search 81

z = 1 0.1 < z < 8 1.0

0.5

0.0

0.5 (J-K)

1.0

1.5

2.0

2.5 2.5 2.0 1.5 1.0 0.5 0.0 (B-u)

Figure 4. Color-Color plot of (J −K) against (B −u) for the initial LAMPSS catalogue. EGOs are separated from stars by spectroscopic redshift (stars: blue, galaxies: red). The black line shows our chosen conditions to remove EGOs (equations7 and8) width, ∆λ, of each filter was found by integrating the profiles with respect to wavelength using the trapezoidal rule. The effective transmission wavelength of each filter was computed using a weighted mean method outlined in the Ap- pendix in §??. The band widths and effective transmission wavelengths of each filter used can be found in Table2. We start by normalising the flux densities, such that all stars appeared to be located at 10 parsecs. To achieve this, firstly, the Stefan-Boltzmann Law, given by,

L = 4πσR2T 4 (11) where L = absolute luminosities, σ = Boltzmann constant, R = Radius of star and T = Temperature of star, was employed to calculate the luminosity given the temperature and radius. The range of radii and temperatures used for each spectral type are shown in Table5. Subsequently, the interpolated stellar spectra were integrated with respect to the wavelength (in angstroms, A)˚ to obtain the total ﬂux Figure 5. Spectrum of a G-type star of temperature T = (an example spectrum can be found in Figure5). The lumi- eﬀ 5750 K and [Fe/H] = 0.0 nosity, L, at 10 pc was found using the following equation:

L = 4πd2F (12) tive transmission from the filter profile, i indicates the ith data-point and R is a ratio of the luminosity obtained via −1 −2 −1 Where F is the total flux in units of erg s cm Hz Stefan-Boltzmann law, Ltypical, and the luminosity at 10 pc 19 and d is 3.086×10 cm (10pc). The convolved flux, Fconv,i, as calculated using Equation 12. Multiplication by the ratio, a function of wavelength was computed using the following R, ensured that integrating the convolved flux would result formula: in the total flux in each filter as seen from 10 pc. This was repeated for each filter. An example of a Solar-like spectrum F = F I R (13) conv,i i i convolved through the i, g, r, v, u and CaHK filters is pre- where F is the flux from the spectrum, I is the rela- sented in Figure6.

NLUAstro 1, 75-90 (2019) 82 Worrell et al. (LAMPSS)

-1.0 O -0.8 B -0.6 A -0.4 F G -0.2 K 0.0 M 0.2 0.4 0

(CaHK-g)-1.5(g-i) 0.6 -1 [Fe/H] 0.8 -2 1.0 -3 1.2 1.4 -4 1.6 -5 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 g-i

Figure 7. Variation of the stars’ colours with metallicity. The shape of data-points represents the spectral type, whereas the colour indicates the metallicity. The hottest stars are shifted to- Figure 6. Solar-like spectrum from Figure5 convolved through wards the negative colour indices along both axes. Here we con- the i, g, r, v, u and CaHK filters. centrate on A, F, G and K type stars, showing the variation of stars’ colour with metallicity. The shape of data-points represents the spectral type, whereas the colour indicates the metallicity. Subsequently, the convolved spectra were integrated us- The hottest stars are shifted towards the negative colour indices ing trapezoidal rule to obtain the total flux seen through along both axes. each filter, Feff . The effective flux was then converted into flux density per unit wavelength, f , as follows: λ standard deviation of the lists. The results were stored in a Feff Table to be later used in TOPCAT to produce magnitude f = (14) λ ∆λ graphs. where ∆λ is the previously calculated filter band width from Table2. This was converted into flux density per unit 4.3 Colour-Colour Predictions frequency, fν , using Equation6 with λ being the effective transmission wavelength, λeff , obtained using Equation ??. In order to be able to distinguish between spectral types the Knowing the flux density, fν , the AB magnitudes in each (g−i) index range for each spectral class was estimated using filter were computed using equation5. Figure7. This was necessary as it allowed the star volume density per spectral type calculations at a later time in the project. The summary of the (g − i) ranges is presented in 4.2 Estimating Stellar Radii Table6. Figure7 shows a colour plot of all stars used for the While the temperatures and metallicities of each star were theoretical predictions. In order to estimate metallicities for known, the radii values were assumed to be the average radii our stars, the y-axis values were chosen to be (HK − g) − for each spectral class and the uncertainties were chosen such 1.5(g − i) colour index and the x-axis values were (g − i). that the smallest and largest stars of each spectral type were This was in agreement with the colour combinations used encompassed by the errors. In order to implement this, a in the Pristine survey (Starkenburg et al. 2017) and this Gaussian distribution of stellar radii was generated and a combination of magnitudes resulted in the largest spread of random radius value, within 3 standard deviations of the data-points corresponding to stars of metallicities [Fe/H] ≤ average radius for each spectral type, was picked to perform −3.0. the luminosity calculation. The summary of radii used is Figure7 presents a close-up concentrating on F, G and presented below in Table5. K type stars and providing a clearer picture of how metallic- For convenience, and because we do not expect O type ity affects the intrinsic colour. The more negative the (g − i) Population II stars, the upper size limit for this spectral type index (x-axis) the hotter the star (the y-axis is inverted). was set to be R = 26.6R . Similarly, since M type Popula- Generally, stars of lower metal content tend to shift towards tion III stars are not expected to be observed the lowest size more negative values along the y-axis within the same spec- limit for M type stars was set to be R = 0.3R . The process tral class. This is in agreement with theory, indicating metal of computing AB magnitudes as described in Subsection 4.1 poor stars appear bluer than their metal rich spectral type was repeated for a 1000 iterations for each stellar spectrum, companions. each time with a different, randomly generated radius. This was to ensure that that magnitude predictions are as accu- rate as possible, despite not having access to any information 4.4 Metallicity Conditions about the stars’ sizes and distances. After each iteration the radii, luminosities and magnitudes in each filter were stored The magnitude predictions from §4.1 allowed us to derive in separate lists. The final values and uncertainties of those metallicity conditions to determine the metallicities of the quantities were computed by calculating the average and the stars in the LAMPSS. This was done by using Numpy’s

NLUAstro 1, 75-90 (2019) LAMPSS: Lancaster Astrophysics Metal Poor Star Search 83

Spectral Type Radius (R ) Temperature Range (K) Metallicity Range O 16.600 ± 10.000 > 30, 000 [Fe/H] = 0.0 B 4.200 ± 2.400 10, 000 − 30, 000 −1.0 ≤ [Fe/H] ≤ 0.0 A 1.600 ± 0.200 7, 500 − 10, 000 −3.0 ≤ [Fe/H] ≤ 0.0 F 1.275 ± 0.125 6, 000 − 7, 500 −5.0 ≤ [Fe/H] ≤ 0.0 G 1.005 ± 0.095 5, 000 − 6, 000 −5.0 ≤ [Fe/H] ≤ 0.0 K 0.830 ± 0.130 3, 600 − 5, 000 −1.0 ≤ [Fe/H] ≤ 0.0 M 0.500 ± 0.200 < 3, 600 [Fe/H] = 0.0

Table 5. The summary of radii and temperatures used for magnitude calculations and ranges of metallicities obtained from the Pollux database for each spectral type.

Spectral Type g − i Condition O <-0.50 B -0.5 to -0.40 1.0 0 A -0.40 to -0.25 -1 F -0.25 to 0.30 0.5 -2 G 0.30 to 0.80 -3 -4 0.0 K 0.80 to 2.40 -5 M >2.40 0.5 Table 6. The g-i conditions used to separate dataset into spectral types. 1.0 (HK-g)-1.5*(g-i) 1.5 polyfit function, fitting each line to the POLLUX data points 2.0 for each metallicity in terms of (g − i). When using different degrees of the fit, it was found that that the POLLUX data 2.5 points at (g − i) ≈ −1 and (g − i) 1.5 greatly affected the & 1.0 0.5 0.0 0.5 1.0 1.5 2.0 fitting of the curves. When these points were removed from (g-i) consideration, each fit used degree 2 in the polyfit function. In order to then create conditions for each metallicity, the intermediate lines of the fitted lines were found such that the Figure 8. Plot showing the conditions created for each metallic- fitted line in between two adjacent intermediate lines would ity over the theoretical POLLUX data. The midlines showing the be the metallicity contained in the two intermediate lines. boundaries for each metallicity are shown in black. Metallicity Figure8 shows the fitted lines in colour and the intermediate conditions for each [Fe/H] step are listed in Table ??. lines in black. Removing points with (g−i) . −1 and (g−i) & 1.5 will not affect our predictions for G type stars because for G type [Fe/H] Number stars in the LAMPSS catalogue 0.3 / (g −i) / 0.8. However 0 1145 our (g − i) range for K type stars lies partly outside of the -1 352 (g − i) range where our metallicity conditions are valid. -2 259 The equations of these lines that allowed the metallic- -3 61 ity conditions to be created are shown in table ??. It must -4 21 -5 92 be noted that the conditions above metallicity 0 and below metallicity -5 have not been created, meaning any star that Table 7. Summary of number of object of each metallicity found falls in the conditions for 0 metallicity may be, for example, before the visual checks with HST images were carried out. This a star with metallicity +1. This is the same for stars in the table includes all spectral types. conditions for metallicity -5, which may have lower metallicity (although this is extremely unlikely). Furthermore, the presence of ’hard borders’ creates limitations because the cut described in section 3.3. It was expected that a large metallicity of each star is only known to be within a region number of objects included in Table7 were galaxies, thus it and does not have a precise value. was decided that it was necessary to visually check images of those objects and remove any unwanted sources using HST images. 4.5 Visual Inspection of the Most Metal-Poor Images taken by the Hubble Space Telescsope (HST) Candidates were downloaded and any objects which were not point-like Table7 shows the number of stars of each metallicity found were rejected. Only objects with [Fe/H] =-3, [Fe/H] =-4 and after applying the metallicity cuts. The number of star with [Fe/H] =-5 were checked, as stars of higher metallcities were [Fe/H] =-3, [Fe/H] =-4, [Fe/H] =-5 appeared to be unex- not the primary focus of this project. An example of a source pectedly large. This was due to the fact that our star sam- which was removed (a galaxy) is shown in Figure9 (left). ple was contaminated with galaxies (30% contamination), Sources of larger metal abundance were not considered. The since it was impossible to remove all galaxies in the single results are summarised in Table7.

NLUAstro 1, 75-90 (2019) 84 Worrell et al. (LAMPSS)

Figure 9. Side by side comparison of two HST frames centered on catalogue objects. Left: Frame centered on a galaxy – evident Figure 10. HK apparent magnitude (300 aperture) histogram for by elongated shape and diffuse edges (LAMPSS 944). Right: the entire LAMPSS catalogue. Counts begin to fall off at mHK ≈ Frame centered on a star - this has a circular, point-like profile 25.3, so this was used as the maximum magnitude for volume (LAMPSS 1254). calculations.

4.6 Distance Calculations Table ?? shows the absolute magnitudes used to calculate the distances. After applying the galaxy cuts and removing any sources Calculating the (g − i) index using the extracted ap- that are to faint, or have unknown magnitudes, in the fil- parent magnitudes, and therefore determining the spectral ters of interest, namely i, g, r, v, u and CaHK, distances type, ensured the code was able to pick out an appropriate to all the remaining objects were calculated. This allowed absolute magnitude. A loop was used to calculate the (g −i) the qualitative assessment of volume density per spectral index, choose an absolute and an apparent magnitude and type and volume density per metallicity. The distance was calculate the distance using equation 15. The uncertainties obtained through the distance modulus equation: were propagated using: √ m − M + 5 σd,j = 10 ln 10 a + b (19) log (d/pc) = (15) 10 5 where 2 Where d is the distance in pc, m is the apparent mag- −Mj mj mj 5 5 5 −1 2 nitude and M is the absolute magnitude. The distance was a = 10 2 5 σmj (20) calculated 5 times, each time using magnitudes in a different filter. The CaHK band was not used for this step. Since m −M −M 2 it is sensitive to metallicity, the magnitudes and therefore j j j −1 2 b = 10 5 2 5 5 5 σ (21) distances are unreliable. The apparent magnitudes and their Mj corresponding errors were extracted from the catalogue. The Here, σ is the distance uncertainty, M refers to the absolute absolute magnitude in each filter was determined using the- d AB magnitude, m represents the apparent AB magnitude oretical data from subsection 4.1. Firstly, the script sorted and the subscript j refers to the jth filter. This process was the data into spectral types based on temperature given in repeated 5 times, each corresponding to a different filter. Table5 and calculated an average AB magnitude in each The final distance value was the mean of those 5 computa- filter for each spectral type. This average value had an er- tions, whereas the final uncertainty was obtained using the ror equal to the standard deviation of each point where the standard deviation formula (Equation 16). standard deviation was given by: ! 1 PN (M − M)2 2 σ = i i (16) 4.7 Calculating Volume SD N 2 − N Volumes for each spectral type of star were calculated by where N was number of values, M was the mean absolute approximating our region of sky as a section of a sphere. th magnitude and Mi was the i magnitude used for the av- This was a valid approach, as the solid angle subtended 2 erage calculation. Each data point also had an associated by the spherical section was found to be only 1.010deg 2 error with its band magnitude which was incorporated into (≈41253deg in a sphere) by studying the RA and Dec of this standard deviation. The combined error was calculated our un-cut catalogue. The radii of the sections was calcu- using: lated by using HK magnitudes of stars of solar metallicity obtained from data by methods described in sec- 1 POLLUX ! 2 tion 4.1 and the distance modulus (Equation 15). The max- 1 X 2 σ = σ (17) CB N j imum mHK was assumed to be 25.3 by judging where the j counts of Figure 10 began to fall off. The results of the vol- th ume calculations can be seen in Table8, B stars were not where σj is the j error in a particular filter. Therefore the total error of each average is given by: included as none were present in the catalogue after applying the star-galaxy seperation criteria (§3.3) and the (g − i) 1 2 2 2 σ = (σSD + σCB) (18) conditions (§4.3).

NLUAstro 1, 75-90 (2019) LAMPSS: Lancaster Astrophysics Metal Poor Star Search 85

3 Spectral Type mHK dmax[kpc] V [pc ] 350 O 4.512 143.7 3.043x1011 A 4.815 125.0 2.003x1011 300 F 5.407 95.19 8.846x1010 250 10 G 6.186 66.49 3.105x10 200 K 10.299 10.00 1.026x108 Count 150 M 10.806 7.921 5.097x106 100 Table 8. Absolute HK magnitudes (m ) for stars of solar metal- HK 50 licity at the central temperature of each spectral type, obtained 0 from POLLUX data. Maximum distance, and volume for each 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 spectral type are also given. B stars are disregarded as we found distance (pc) x105 none within our catalogue. Figure 12. Histogram presenting the number of stars per dis- Spectral Type Number tance interval. This shows the halo limit where the number of stars drops off. O 0 B 0 A 2 F 131 largest set being K-type stars, F-type stars and just two G 1064 A-type stars. There were no stars identified that were of K 575 spectral types O, B or M. These values of each spectral type M 0 should be expected as G and K type stars are two of the most Table 9. The finalised number of stars of each spectral type common spectral types due to having such long lifetimes and remaining after applying the galaxy cuts (using the J − K mag- COSMOS does not track star-forming regions where O and nitude and Hubble and removing the stars with [Fe/H] >-3.0. B type stars are formed, hence none of them being identified. The lack of M-type stars can be explained by the fact they are the dimmest, thus hard to detect at large distances. This agrees with our theory as we expected most of the stars to A F G K be G type, and a lesser amount to be K type. The number of F types was surprising seeing as they should only live 3- 5 8 Gyrs, which is less than the lifetime of the Universe. This implies these stars must have been created in a metal-poor 6

] region of the galaxy between 3-8 Gyrs ago, and then moved 3 c

p out into the halo due gravitational interactions. [

) 7 V

/ Figure 14 shows a 3D positional plot of the LAMPSS N

= catalogue stars. Hotter stars, A- and F-types for our sample,

( 8 0

1 can be observed at much further distances than their cooler g

o [Fe/H]=0 l partners due to their higher luminosities. 9 [Fe/H]=-1 [Fe/H]=-2 [Fe/H]=-3 10 [Fe/H]=-4 [Fe/H]=-5 5.2 Distance to Halo

0.5 0.0 0.5 1.0 1.5 2.0 2.5 (g - i) Since the project focuses on metal-poor stars in the halo, it is important to have an estimate of where the halo is located. The plot of the distribution of distances, which can be found Figure 11. The number density of each spectral type per unit in Figure 12, shows there is a drop off in the number of volume per metallicity in the Halo. The density increases with stars at a distance of approximately 120 kpc. This gives an (g − i) and lower metallicity stars are only G and K type stars. approximate distance of the halo because most stars within a galaxy are in the disc and the bulge, with relatively few in the halo. This value fits nicely within previous studies 5 RESULTS to estimate the distance to the edge of the halo of 100 − 5.1 Quantitative Analysis 200 kpc (Battaglia et al.(2005)). Of course, the stars stretch beyond 120 kpc, however the number of them past this value After applying a variety of cuts described in §4 the spectral is significantly reduced. type of all the remaining stars was determined by calculating the (g − i) indices. The number of stars per spectral type can be found in Table9. 5.3 Metal-Poor Candidates Using the maximum distances for stars of different spectral type presented in Table8 and the area of the field of Having manually checked the images of the [Fe/H] = −3, view which was calculated to be 1.010deg2, a graph showing -4 and -5 metallicity candidates, the number of stars per the number density per spectral type was plotted. This can metallicity was significantly reduced. The finalised number be found in Figure 11. of candidates of each metallicity per spectral type is shown Table9 shows that of the stars found, a significant main Table 10. jority of the stars identified were G-type stars, with the next When the conditions for each metallicity (see Table

NLUAstro 1, 75-90 (2019) 86 Worrell et al. (LAMPSS)

Number of Stars in Spectral Type Metallicity 0 G K A F Total

0 886 128 2 129 1145 -1 -1 142 208 0 2 352 -2 34 225 0 0 259 400000 -2 -3 1 14 0 0 15 [Fe/H] -4 0 0 0 0 0 -3 -5 1 0 0 0 1 200000 distance (pc) 2.75 2.5 Table 10. The diﬀerent numbers of spectral types at diﬀerent 2.25 0 2 -4 149.5 DEC J2000 metallicities found after removing galaxies using Hubble images. 150 RA J2000 150.5 -5

0 0.5 Figure 14. Distance plot showing the stars in the LAMPSS catalogue after magnitude and galaxy cuts. The auxiliary axis rep-

0.0 resents the metallicity. 1

0.5 tive values along the x-axis and more negative values along 2 y-axis as star temperature decreases. However, in the K re- 1.0 gion at (g−i) ≈ 1 this is not the case. Here, the y-axis values [Fe/H] start increasing, reaching a maximum of ((HK-g)-1.5(g − i)) (HK-g) - 1.5(g i) 3 1.5 at (g − i) ≈ 1.8 (it is important to remember the y-axis is inverted), before starting to drop again and steadily decreasing through the M region. This is most likely due to a 2.0 4 suddenly increasing ﬂux density in the i-band for the coolest stars as well as the presence of molecules causing molecule 2.5 absorption lines in the spectra of the M-type stars. Hence,

5 it cannot be ascertained where on the (HK-g)-1.5(g − i) 0.0 0.5 1.0 1.5 2.0 (g-i) vs. (g − i) colour-colour plot metal-poor K-type candidates would appear, which reduces the number of candidates with Figure 13. Metallicity plot of finalised catalogue with (g −i) and [Fe/H] = −3.0 to just one (LAMPSS 1586). ((HK −g)−1.5(g −i)) colour indices. Metallicity conditions have The most-metal poor candidate, LAMPSS 1229 (Shown been applied and for objects with [Fe/H] ≤ −3.0 galaxies have in Figure 15), had a (g − i) index of 0.4736 meaning it been visually checked and separated from the catalogue. was classified as a G-type. Having produced reliable metallicity curves in the G region due to the large number of theoretical data-points it was agreed that this is the ??) were applied to the reduced LAMPSS catalogue, as ex- most promising metal poor candidate. The distances to pected, the majority of the stars are defined as zero metal- LAMPSS 1229 and LAMPSS 1586 were computed to be licity stars, with a decreasing number of stars as [Fe/H] de- 78210.7±9368.5 pc and 118348.3±20978.1 pc with an esti- creases (see Figure 13). As explained in section 2.1, this is mated lifetime of 8.29±1.43 Gyrs and 10.86±2.87 Gyrs re- because most stars in the present age formed from an ISM spectively putting them in the halo region of the Milky-Way. with heavier metals than hydrogen and helium. It is important to notice that it was realised that the Only G and K type stars were found to have [Fe/H] ≤ CaHK filter profile was not normalised before calculating −3.0. The colour-colour metallicity diagram for G type stars the magnitudes - instead its peak transmission was at ≈0.9. is shown in Figure 13, while a similar plot for K type stars is The effect of this was checked by carrying out the normal- shown in Figure ??. A detailed summary of stars surviving isation and re-calculating the CaHK magnitudes from our those metallicity cuts is shown in Table 11. theoretical spectra. It was found that magnitudes agreed While the large number of stars with [Fe/H] ≤-3 is sur- to 2 decimal places, thus it was not necessary to re-plot the prising as it would appear to contradict the assumption that metallicity criteria as the effect was negligible. Furthermore, most metal-poor stars will most likely be G-type stars, it is since the CaHK filter was not normalised to 1, the fluxes crucial to emphasise that we are uncertain about the metallicities of all candidates classified as K-type. Since theoretical spectral data obtained contained no K-type stars of metal- 5.4 Metallicity Distribution Function (MDF) of licity lower than [Fe/H] =0 it cannot be confirmed with any the Halo degree of confidence that these particular sources are metal- poor. When fitting the metallicity criteria shown in Figure8, To investigate the metallicities of stars in the halo, a metal- the [Fe/H] =-3.0 curve in the K region (0.8 ≤ (g −i) ≤ 2.40) licity distribution function of the halo has been produced, was an extrapolation of data-points of all preceding spectral which can be found in figure 16. Since we are concerned types. Moreover there is unusual positioning of the K and with stars located towards the halo, the minimum distance M data points, which does not appear to follow a pattern thre shold for this histogram was set at being at a distance which the other spectral types do. The points corresponding greater than 78 kpc, so as to include the singular -5 metallic- to different metallicities tend to shift towards more posi- ity star identified (which is at a distance of 78.2 kpc). Due to

NLUAstro 1, 75-90 (2019) LAMPSS: Lancaster Astrophysics Metal Poor Star Search 87

ID [Fe/H] (g-i) Spectral Type Distance (kpc) RA (J2000) DEC (J2000) Lifetime (Gyrs) LAMPSS 1229 -5 0.4736 G 78±9 149.90224 2.327138 8.29 ±1.43 LAMPSS 1875 -3 1.3762 K 32±8 150.6464 2.295457 104.99 ± 38.42 LAMPSS 1671 -3 1.8790 K 12±7 149.7748 2.0275638 189.45 ± 168.81 LAMPSS 1586 -3 0.5746 G 118±21 149.90584 1.8717754 10.86 ± 2.87 LAMPSS 1503 -3 1.1630 K 23±6 150.09969 1.890762 80.38 ± 30.80 LAMPSS 1457 -3 1.3449 K 26±7 150.18236 1.9493028 116.32 ± 43.88 LAMPSS 1417 -3 1.8608 K 20±10 150.29182 1.9552598 192.88 ± 158.51 LAMPSS 1373 -3 1.2362 K 20±5 150.40768 1.8676664 92.88 ± 32.19 LAMPSS 1254 -3 1.5835 K 22±11 149.73572 2.1850543 162.76 ± 118.50 LAMPSS 1111 -3 1.1107 K 35±12 150.05904 2.295457 67.87 ± 34.36 LAMPSS 1004 -3 1.8913 K 18±9 150.3845 2.2432206 171.97 ± 122.68 LAMPSS 627 -3 1.2060 K 27±9 150.36208 2.5158165 108.13 ± 52.31 LAMPSS 392 -3 1.6818 K 20±10 149.87767 2.78925 172.81 ± 123.55 LAMPSS 260 -3 1.7448 K 15±4 150.13509 2.6929514 129.87 ± 52.52 LAMPSS 239 -3 1.5282 K 20±10 150.19594 2.824687 130.72 ± 79.47 LAMPSS 56 -3 1.7817 K 22±12 150.5649 2.766089 187.93 ± 147.01

Table 11. Metallicities estimated using metallicity criteria summarised in Table ?? and spectral types of our candidates deduced through calculating the (g − i) index and comparing it to the predictions summarised in Table6.

1100

1000

900

800

700

600 Count 500

400

300

200

100 0 -5.0 -4.5 -4.0 -3.5 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 [Fe/H]

Figure 16. Distribution of stellar metallicities in the Milky-Way halo. It is clear as the number of stars decreases with decreasing metallicity, which is in agreement with the theory.

the total number of metal-poor stars found in the estimated halo. It was found that ≈ 613, 000 stars with [Fe/H]<-3 and Figure 15. HST image of object LAMPSS 1229 ≈ 40800 stars with [Fe/H]<-5. These numbers are consider- ably lower than those presented on M. Magg et al.(2019), this is likely because there is a signiﬁcant amount of extrap- the large number of higher metallicity stars, this star does olation used in reaching these estimates so the error involved feature on the histogram. will be large.

6.2 Halo MDF Compared with Data from HK 6 DISCUSSION Survey 6.1 Number of Metal-Poor Stars in Halo A previously produced metallicity distribution function Using the value of 78 kpc as the starting point of the halo (at (MDF) of the halo of the Milky Way using data from the HK which the disc is not included and the halo dominates) and Survey (Beers & Christlieb 2005) can be used as comparison 144 kpc as the edge of the halo (which ﬁts with the value of to the LAMPSS MDF (Figure ??). It is noted that the area 130 kpc from Battaglia et al.(2005)), an estimate of the vol- surveyed for this distribution is much greater (≈ 6000 deg2 ume of the halo of the Milky Way was found, and hence the vs ≈ 1 deg2), however the HK survey only probes to a mag- fraction of this that has been surveyed for the data. With nitude of ≈ 15.5 compared to ≈ 26 meaning the data used this fraction known, and the known number of metal-poor for this report extends to a much deeper ﬁeld of view. candidates that came out of the this study, an estimate of Even with these dissimilarities, a comparison of the

NLUAstro 1, 75-90 (2019) 88 Worrell et al. (LAMPSS)

LAMPSS MDF to the Beers MDF shows that, as expected, 0.0 the majority of stars have Solar-like, or close to Solar-like -0.5 metallicities. Beers’ MDF shows a peak in metallicity at ≈ -0.5, whereas for the LAMPSS MDF the peak is centred -1.0 around metallicity of zero, however since the LAMPSS cuts -1.5 are produced in increments of one, it is not possible to determine if a peak would appear around a half-integer value. -2.0 [Fe/H] (WARP) Moreover, Figure ?? appears to have another peak around -2.5 ≈-2.2, which does not appear in the LAMPSS MDF. This is also due to the fact that the methods used to estimate metal- -3.0 licities in this project meant it was only possible to approx- -3.5 imate the metallicities as integer values. A more complex -4.0 method of computing metallicities would be necessary to -3.0 -2.8 -2.6 -2.4 -2.2 -2.0 -1.8 -1.6 -1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 [Fe/H] (LAMPSS) quantitatively compare both MDFs. Nevertheless, the gen- eral trends of both MDFs agree - as metallicity decreases so does the number of stars in agreement with the assump- Figure 17. WARP (Jenkins et al. 2019) metallicities vs. tion that metal-rich stars are more abundant in the Universe LAMPSS metallicities of ﬁnal catalogue objects. today.

A F G K 6.3 Pristine 4 [Fe/H]=0 [Fe/H]=-1 The Pristine survey (Starkenburg et al. 2017) uses a very [Fe/H]=-2 [Fe/H]=-3 similar method to this project, by using Ca HK lines with 6 [Fe/H]=-4 the wide-ﬁeld imager MegaCam on the CFHT to search for [Fe/H]=-5 ] 3

[Fe/H] = −0.5 down to the extremely metal-poor [Fe/H] < c 8 p [

2 )

−3.0 stars in an area of over 1000 deg of the galactic V / N halo (with an accuracy of 0.2 metallicity). They had a = 10 ( 0 1

success rate of 24% of selected and filtered stars being g o [Fe/H] < −3.0, and 85% of the remaining candidates still l being reasonably metal poor at [Fe/H] < −2.0, from a total 12 of initially 17500 stars, that were filtered down to a total of 7673 stars. In comparison with the results obtained in this 14 project in which 15 stars were determined to have a metal- 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.5 licity of [Fe/H] ≈ −3, which is approximately 0.84% of our (g - i) filtered stars, it can be seen that the surveys are in support of each other in the discovery of potential EMP candidates. Whilst the LAMPSS survey only swept an area of 1deg2 Figure 18. LAMPSS number densities plotted with WARP val- it was still deduced that a reasonably significant amount of ues (shown as N; Jenkins et al. 2019). WARP data gives smaller the determined stars would fall in the metal-poor category. densities over all spectral types. However Pristine was able to deduce which potential candidates were white dwarfs, as white dwarfs typically show no Firstly, while WARP implemented a method of calculating Ca absorption features they are a potential source of con- a metallicity that included decimal error estimations (see tamination in the metal-poor samples, unlike the LAMPSS Jenkins et al. 2019), our process only allowed the estimation catalogue in which we were unable to determine the dif- of [Fe/H] to an integer precision. Furthermore, both groups ference of white dwarfs and variable stars from the filtered did not carry out visual checks of images for metal-rich stars samples, causing a potential contamination error. thus rejecting galaxies. A result of this can be seen in the large spread of data-points, particularly around [Fe/H] = 0.0 and [Fe/H] = −1.0 (LAMPSS). 6.4 Comparison to WARP (Jenkins et al. 2019) WARP (We Are Really Poor; Jenkins et al. 2019), have a 6.5 Calcium Abundances somewhat parallel effort to ours to find Population III candidates, however they use data from the Isaac Newton Tele- For these results, we have assumed that the Ca II lines and scope (INT). therefore calcium abundances trace those of of iron, as a Final WARP and LAMPSS catalogues were matched means to produce the metallicity of a star. However studies by the RA and Dec. of the objects, detecting a total of have shown that over large metallicities differences, the ho- 42 astronomical sources. Figure 17 shows the linear corre- mogeneity of Ca II to iron does not remain. Vásquez et al. lation between metallicities computed by independently by (2015) calibrated their [Fe/H] using K giant stars in the WARP and LAMPSS for those objects and it indicates that, Galactic halo, finding that by adding a second-order correc- on average, metallicity estimates from both groups agree. tion could get a linear correlation over −2.3 < [Fe/H] < 0.7, The maximum allowed error in arcseconds was set to 1 arc- however without, a range of −1 < [Fe/H] < 0.7. Carrera sesond. This plot was greatly limited by a number of factors. et al.(2007) also recorded similar results, finding linear cor-

NLUAstro 1, 75-90 (2019) LAMPSS: Lancaster Astrophysics Metal Poor Star Search 89 relation over −2.2 < [Fe/H] < 0.47, using almost 500 stars cause LAMPSS 1229 is a sub-giant, so has a more negative in 29 galactic open and globular clusters. The only con- apparent magnitude than calculated, therefore Equation ?? cern is that we did not use the Ca II triplet, we used the cannot be accurately used to determine its lifetime. CaHK lines, but seeing as both lines are from singly ion- Several limitations affected the project. Firstly, when ized calcium, these results should still apply. As in our data working out the magnitude cuts as shown in Figure3, we (§4.3), we changed the metallicity data into g-i magnitudes found a value of mHK = 26, however the actual drop-off as metals affect g-i values. Therefore there might be a range magnitude was mHK = 25.3. Secondly, when manually re- where cannot be certain of the metallicity when using CaHK. moving galaxies shown in §4.5, some galaxies would be be- However both these studies were performed on giant stars, hind the stars affecting their emission data, we therefore whereas we are looking at main sequence stars. When a star could not use this star. This reduces our completeness, as of mass in the range 0.8M < M < 5M exhausts its core some stars could not be included in our sample. The condi- of Hydrogen, hydrogen shell burning occurs around the inert tions created to determine metallicities were created using core. Once the star reaches the RGB (Red Giant Branch), a main sequence star model. This means that all results the star begins to ‘dredge’ up its fusion products. This hap- acquired are only valid for main sequence stars and further pens as the convective zone begins to expand further down processes would need to be carried out to come up with esti- into the star, bringing down fusion materials and take up mates for non-main sequence stars. This is explained further fusion products. If the dredge up brings up fusion products in §8. then the metallicity of the giant will increase, or depending Additionally, the CaHK filter was not normalised (see on the makeup of the star, might also reduce it. This may Figure2). However, the difference in area between the nor- explain the curve in Ca observed in (Vásquez et al. 2015) malised and un-normalised was such that it only affected and (Carrera et al. 2007). Main sequence srars are more con- calculated magnitudes by a factor of a hundredth. If any- sistent than this, therefore this may not apply to our stars. thing, this error would lead to conservative [Fe/H] estimates. Therefore, due to time constraints, and the minimal effect on our EMP star sample, we chose not to correct this. Overall, we present a sample of potentially EMP stars 7 CONCLUSION AND FUTURE WORK and one UMP star within the stellar halo obtained by We present the results from the Lancaster Astrophysics analysing CaHK lines. We find a correlation between our Metal Poor Star Search (LAMPSS) to identify the most results and the parallel study conducted by WARP (Jenk- metal-poor stars towards the galactic halo using the ins et al. 2019). All metal-poor candidates identified require metallicity-sensitive CaHK lines. The project was con- spectroscopic follow up, especially due to the uncertainties ducted using observations from the CaHK filter of the in the true metallicities of all K type LAMPSS stars discov- MegaPrime/MegaCam on the CFHT, in combination with ered from extrapolation of metallicity conditions. the broad-band optical and near infrared photomotery of 2 COSMOS. We surveyed 1.010 deg of sky to mHK ∼ 26, approximately 5 orders of magnitude deeper than the Pristine 8 IMPROVEMENTS survey (Starkenburg et al. 2017). 1,772 stars were retained from the initial LAMPSS cat- In our methodology, we concentrated on Calcium II absorp- alogue of 62,652 sources after removing entries with undeter- tion lines, whereas other surveys such as Starkenburg et al. mined u, g, r, i, z or mHK and removing galaxies (with 70% (2018) used Na I (Sodium), Mg I (Magnesium), Si I (Sili- completion and 30% contamination). Colour conditions for con), Ca I (Calcium), Fe I (Iron) and Ti II (singly ionised different [Fe/H] values were found by creating metallicity- Titanium). Using more element absorption lines means we sensitive colour-colour diagrams using a range of data from could be more confident in our metallicity data, as we would POLLUX. From these conditions, 16 stars of [Fe/H] . -3.0 expect the abundance of any metals to increase the metallic- were identified. ity. Therefore in a metal-poor star, all these elements should One star, LAMPSS 1229 (see Figure 15), was found to be of little abundance rather than just one. Equally knowing have [Fe/H] ∼ -5.0. LAMPSS 1229 falls in the (g − i) bin the carbon abundance in a star would help avoid situations of G type stars and lies within the stellar halo at a distance such as the one described in §2.2. from Earth of 78 ± 9 kpc. These sources were confirmed as Furthermore, spectroscopic data would allow the stars by manually checking HST imagery for each RA and lithium abundance of proposed metal-poor stars to be de- Dec. A second G type star, LAMPSS 1586, was found to termined, a particularly useful piece of information because have a metallicity of [Fe/H] ∼ -3.0. It was found to have a stars with a metallicity [Fe/H] < −3 will contain an amount distance of 120± 20 kpc, just within in the stellar Halo. The of lithium below the ‘Spite Plateau’, a baseline of constant lifetimes of LAMPSS 1229 and LAMPSS 1586, determined lithium abundance of ≈ 2.26 (Fu et al. 2015) which features from their distances and apparent magnitude (Equation ??), in less evolved, higher metallicity stars ([Fe/H] ≤ 3). is given by 8.29 ± 1.43 Gyrs and 10.86 ± 2.87 Gyrs respec- In addition, we assumed that all stars observed were in tively. This is surprising as LAMPSS 1586 has a longer life- their main sequence, however in reality some of these stars time than LAMPS 1229, even though LAMPSS 1586 has a may be subgiants, but without knowing the radius of the metallicity of -3.0 compared to -5.0. A lower metallicity star stars, we cannot determine whether this is true. If the ra- should have a greater lifetime if it was created in the early dius was known, instead of having to estimate it, we could Universe. For LAMPSS 1229, we would expect a lifetime plot a HR diagram and determine the stars stage in its life- > 11 Gyrs (Starkenburg et al. 2018), however its approxi- time. A star on main sequence at of 1.0 kpc of radius 1.0 R mated lifetime was far less than this. This is probably be- at Temperature of 5000 K, will therefore look like a subgiant

NLUAstro 1, 75-90 (2019) 90 Worrell et al. (LAMPSS) of radius 2.0 R at a distance of 2.0 kpc as their magnitudes Jones E., Oliphant T., Peterson P., et al., 2001, SciPy: Open for g, for example, will be the same. This will also change source scientific tools for Python, http://www.scipy.org/ the metallicity of the star, as stars with masses between 0.8 Keller S. C., et al., 2014, Nature, 506, 463 Koh D., Wise J. H., 2018, MNRAS, 474, 3817 M and 8 M (K and G type stars) have non-convective cores, therefore during their SubGiant Branch (SGB) the Laughlin G., Bodenheimer P., Adams F. C., 1997, ApJ, 482, 420 LemaˆıtreG., 1927, Annales de la SociétéScientifique de Bruxelles, core has exhausts itself of hydrogen, and begins Hydrogen 47, 49 shell burning. The next step in this stars evolution will be M. Magg Klessen R., Glover S., H. L., 2019, the first dredge up, which will cause fusion products in the Macaulay E., et al., 2018, arXiv e-prints, core to be brought to the surface, thus increasing the metal- Magg M., Hartwig T., Agarwal B., Frebel A., Glover S. C. O., licity of the star. This is a problem as stars that have low Griffen B. F., Klessen R. S., 2018, MNRAS, 473, 5308 metallicities (-5, -4 etc) will appear as higher metallicities Nordlander T., Amarsi A. M., Lind K., Asplund M., Barklem (-2, -1 etc), and therefore will be ignored. P. S., Casey A. R., Collet R., Leenaarts J., 2017, A&A, 597, A6 Oke J. B., 1974, ApJS, 27, 21 Oke J. B., Gunn J. E., 1983, ApJ, 266, 713 ACKNOWLEDGMENTS Placco V. M., Beers T. C., Reggiani H., Meléndez J., 2016a, ApJ, 829, L24 The authors are thankful to CFHT for performing the ob- Placco V. M., et al., 2016b, ApJ, 833, 21 servations, the COSMOS survey for resulting in such a Ryan S., Norton A., 2010, Stellar Evolution and Nucleosynthesis. rich photometric set of data and David Sobral for his con- Cambridge University Press, https://books.google.co.uk/ siderable guidance throughout the project. This work is books?id=PE4yGiU-JyEC based in part on data from the POLLUX database(http:// Sakari C. M., et al., 2018, ApJ, 868, 110 pollux.graal.univ-montp2.fr) operated at LUPM (Uni- Sarmento R., Scannapieco E., Cohen S., 2018, ApJ, 854, 75 versitéMontpellier - CNRS, France with the support of the Schaerer D., 2002, A&A, 382, 28 PNPS and INSU. We have benefited immensely from the Scharping N., 2017, D-brief public available programming language Python, including Scoville N., et al., 2007, ApJS, 172, 1 NumPy & SciPy (Van Der Walt et al. 2011; Jones et al. Sestito F., et al., 2019, MNRAS, 484, 2166 Stacy A., Bromm V., Lee A. T., 2016, MNRAS, 462, 1307 2001), Matplotlib (Hunter 2007), Astropy (Astropy Col- Starkenburg E., et al., 2017, MNRAS, 471, 2587 laboration et al. 2013) and the Topcat analysis program Starkenburg E., et al., 2018, MNRAS, 481, 3838 (Taylor 2013). This research has also made use of the VizieR Taylor M., 2013, Starlink User Note, 253 catalogue access tool, CDS, Strasbourg, France. We are also Trenti M., Padoan P., Jimenez R., 2015, arXiv e-prints, pp 1–5 grateful for all of Lancaster University’s on-campus caffeine Van Der Walt S., Colbert S. C., Varoquaux G., 2011, Computing dispensaries. in Science & Engineering, 13, 22 Vásquez S., Zoccali M., Hill V., Gonzalez O. A., Saviane I., Re- jkuba M., Battaglia G., 2015, A&A, 580, A121 Weidner C., Vink J. S., 2010, A&A, 524, A98 REFERENCES

Arnould M., Goriely S., Takahashi K., 2007, Physics Reports, 450 This paper has been typeset from a TEX/LATEX file prepared by Astropy Collaboration et al., 2013, A&A, 558, A33 the author. Battaglia G., et al., 2005, MNRAS, 364, 433 Becerra F., Greif T. H., Springel V., Hernquist L. E., 2015, MN- RAS, 446, 2380 Beers T. C., Christlieb N., 2005, ARA&A, 43, 531 Bromm V., Coppi P. S., Larson R. B., 1999, ApJ, 527, L5 Bromm V., Coppi P. S., Larson R. B., 2002, The Astrophysical Journal, 564, 23 Caffau E., et al., 2011, Nature, 477, 67 Capak P., et al., 2007, ApJS, 172, 99 Carrera R., Gallart C., Pancino E., Zinn R., 2007, AJ, 134, 1298 Chiti A., Frebel A., 2019, arXiv e-prints, Forbes D. A., et al., 2018, arXiv e-prints, pp 4–6 Fu X., Bressan A., Molaro P., Marigo P., 2015, MNRAS, 452, 3256 Griffen B. F., Dooley G. A., Ji A. P., O’Shea B. W., Gómez F. A., Frebel A., 2018, MNRAS, 474, 443 Hansen T. T., et al., 2018, ApJ, 858, 92 Hartwig T., Bromm V., Klessen R. S., Glover S. C. O., 2015, MNRAS, 447, 3892 Hartwig T., Bromm V., Loeb A., 2018, MNRAS, 479, 2202 Hubble E., 1929, Proceedings of the National Academy of Science, 15, 168 Hunter J. D., 2007, Computing In Science & Engineering, 9, 90 Ishiyama T., Sudo K., Yokoi S., Hasegawa K., Tominaga N., Susa H., 2016, ApJ, 826, 9 Jenkins A., Wragg F., Butterworth J., Edwards J., O’Neill C., Gough J., Sobral D., 2019, NLUAstro, 1, 21

NLUAstro 1, 75-90 (2019)